SYSTEM AND METHODS FOR PERFORMING MEDICAL PHYSICS CALCULATION
A method of calculating radiation fluence and energy deposition distributions on a networked virtual computational cluster is presented. With this method, complex Monte Carlo simulations that require expansive equipment, personnel, and financial resources can be done efficiently and inexpensively by hospitals and clinics requiring radiation therapy dose calculations.
The present application claims priority to U.S. Provisional Patent Application No. 61/343,419 filed Apr. 28, 2010 and 61/339,711 file Jul. 16, 2010, the disclosures of which are hereby incorporated by reference in their entirety.
STATEMENT CONCERNING FEDERALLY SPONSORED RESEARCH
This invention was made with U.S. government support, National Science Foundation CBET 0853157. The U.S. government may have certain rights in the invention.
FIELD OF THE INVENTION
The invention relates generally to cloud computing or utilization of computational resources on demand via a computer network, and particularly to a system and methods for performing medical physics calculations such as energy deposition and radiation fluence. Such calculations may be used in planning radiation therapy treatments, design and quality control of diagnostic medical imaging, and design of radiation shielding, among other applications.
BACKGROUND OF THE INVENTION
Radiation therapy (or radiotherapy) is commonly used in the treatment of various cancers to control the spread of malignant cells. Radiotherapy may be used for curative or adjuvant treatment. When selecting a dose, many other factors are considered by radiation oncologists, including whether the patient is receiving chemotherapy, patient comorbidities, whether radiation therapy is being administered before or after surgery, and the degree of success of surgery. The amount of radiation used in photon radiation therapy varies depending on the type and stage of cancer being treated.
Delivery parameters of a prescribed dose are determined during treatment planning, part of dosimetry. Treatment planning is currently usually performed on dedicated computers using specialized treatment planning software. The planner will try to design a plan that delivers a prescription dose to the tumor and minimizes dose to surrounding healthy tissues. Depending on the radiation delivery method, several angles or sources may be used to sum to the total necessary dose. Proper dosing remains one of the most important considerations of radiation therapy. With proper dosing, ill cells may be targeted and while healthy cells are preserved.
Dosimetric calculations are a crucial component in the accurate delivery and assessment of radiation therapy. As radiation therapy and diagnostics imaging technologies have become more complex, the associated physics calculations have become more resource intensive. These requirements have been largely met by the exponential increase in processor speed and RAM size, but sometimes outstrip the pace of computer technology even for conventional, deterministic calculation techniques. For example, TomoTherapy, Inc.'s TomoHD ships with a 14 node calculation cluster. Non-deterministic algorithms, such as the Monte Carlo method, demand even greater computing resources than conventional algorithms, but generally offer superior dose calculation accuracy. This is particularly true for complex, heterogeneous treatment scenarios such as particle therapy treatment planning. The use of Monte Carlo simulations is highly desirable for dose calculation in radiation therapy as part of treatment planning or verification. In the recent years, Monte Carlo simulations brought the much-needed dosing accuracy to cancer patients. Monte Carlo simulations represent the gold standard in radiation dose calculation since they include the real physics of the interactions of photons with materials.
Notwithstanding its great results, Monte Carlo simulations have not yet been put into routine clinical use due to long calculation times. For example, it has been reported that times of more than 100 CPU hours are required to simulate proton beam treatment plans when using approximately 2×107 primary protons per field. Thus, while Monte Carlo techniques are widely seen as the gold standard of radiation dose calculations, they are only sparingly used clinically in favor of faster, less resource intensive algorithms at the cost of dosimetric accuracy. The primary barrier to widespread adoption of Monte Carlo techniques has been the requirement of large computing resources to achieve clinically relevant run times, particularly in particle therapy applications. These resources, usually in the form of a computing cluster, require a sizable infrastructure investment as well as associated utility, maintenance, upgrades, and personnel costs. These costs make full, analog Monte Carlo methods effectively unfeasible for routine clinical use. This is especially true in particle therapy.
Large-scale, full Monte Carlo simulations and other resource intensive algorithms are often simply considered unfeasible for clinical settings. Very few, if any, clinics are able to use Monte Carlo simulations for routine verifications of their proton therapy plans. Each simulation may take as many as 240 CPU hours for tens of millions of primaries. The majority of the clinical dosimetry systems are based on faster, less resource intensive algorithms, at the cost of dosimetry accuracy.
Cloud computing is a name given to a set of technologies offered as services over the Internet. Pricing is usually based on the pay-as-you-go model, generally billed in hourly increments, and without set contract periods. This scheme allows cloud services to offer on-demand computing infrastructure, sized to fit the user's monetary needs. Cloud computing has become feasible because of the economies of scale afforded by the commoditization of computer hardware, extensive availability of high bandwidth networks, and growth of free, open source software, including operating systems, such as Linux, and virtual machine software.
For clinical usage, cloud computing has many advantages. Cloud resources can be scaled to meet patient and physics quality assurance demand as it fluctuates on a daily basis. Typical computing clusters often face bursts of usage where they are under-utilized most of the day and night and over-queued at peak periods. The cloud paradigm is particularly well suited for one-off calculations, such as machine commissioning and vault shielding calculations, for which a very large cluster might be desirable, but expanding to even a small cluster would be prohibitively expensive for a single run. Also, hardware upgrade and maintenance is taken care of by the provider, rather than by the user.
It would thus be desirable to find a way to improve dosing accuracy, make Monte Carlo calculations more feasible and accessible to clinics and hospitals, and to reduce the costs and computation times associated therewith. Monte Carlo calculations would be particularly well suited to cloud style distributed computing. This is so because the primary particle histories are completely independent of one another, requiring no communication between processes. Monte Carlo calculations, while parallel, need not maintain data or timing synchronization during execution. Until now, however, Monte Carlo calculations in a cloud environment have not been applied to medical physics calculations. Grid computing has been implemented in medical physics calculations, but it is plagued by the same disadvantages as cluster computing. It is inefficient, expensive, and less accessible for medical facilities, despite being widely considered a forerunner to cloud computing.
There is a need to perform medical physics calculations using virtual computer resources versus local dedicated hardware. The present invention satisfies this demand.
SUMMARY OF THE INVENTION
The present invention calculates energy deposition distributions and radiation fluence on a networked virtual computational cluster. Radiation therapy calculations can be performed using a cloud-based, virtual cluster. The systems and methods disclosed herein allow easy, economical and secure access to high performance computing resources for routine clinical applications. What is disclosed is a cloud computing based distributed Monte Carlo dose calculations system and method for implementing Monte Carlo dose calculations for routine clinical use on such systems. Cloud computing fundamentally alters the economics of such calculations by providing relatively cheap, on-demand, pay-as-you-go computing resources over the Internet. Therefore, very large scale Monte Carlo calculations can be routinely performed by clinics and researchers in an efficient and economical way.
The impetus behind using the cloud computing model as a computing cluster is the desire for large scale processing power without an associated large price tag. Costs associated with the cloud are generally only incurred on a usage basis, whereas in-house hardware incurs capital and maintenance costs. Furthermore, additional potential costs in a typical cluster setting involve personnel, utility, equipment housing, insurance, vendor services, and other costs that might be associated with an in-house model. Generally, a typical 100 node cluster would cost $160,000 USD over a 3 year lifetime, when taking into account approximately $1000 USD per node plus $200 USD in maintenance costs per year per node.
The cost of using a cloud-based cluster is primarily determined by the number of CPU-hours used. It is widely accepted in the industry that each patient needs approximately 100 CPU hours of cluster time. Assuming that a typical patient throughput for a clinic is 1000 patients per year, the CPU time necessary would be approximately 100,000 CPU hours. While the rates vary among the different available providers, the average cost is about $0.10 USD per CPU hour, thus costing a typical clinic around $10,000 USD per year. Additional costs may be added for data storage and transfer to and from the cloud. Such additional costs may be around 5% of the annual CPU costs, resulting in a yearly expense of around $10,500—less than 20% of the estimated annual in-house cluster cost. Moreover, these in-house estimates do not take into account the additional expenses incurred for installation and maintenance of cooling systems typically associated with computing clusters and the power consumption of such computing clusters.
A major concern when transferring or storing personal or patient data on networked hardware is the security and privacy of the data. This may even be regulated by the government. But the nature of the calculations discussed herein—radiation dosing—does not require the transfer of full medical records. In fact, very little personal identification information is needed to be sent over a network, and much of the information can be redacted. For example, names can be replaced with hash tags. Moreover, the population has seemingly demonstrated a certain level of comfort and trust in currently available security implementations, as can be seen from the widely accepted use of cloud-based software for PACSn (electronic picture archiving and communication systems) and record storage. Furthermore, while cloud computing is most often associated with CPU-based computing, it can be deployed with a mix of CPU and GPU (graphics processing unit) resources to provide the best possible results. Still further, such a framework can be used with various portals for easy upload of the Monte Carlo calculations. Also, inside the cloud, all virtual nodes may be booted from pre-stored machine images.
In one embodiment, distributions of energy deposition and fluence due to ionizing radiation are determined using a virtual computer cluster over a network such as a distributed processing framework. The distributed processing framework may dynamically monitor the virtual computer cluster, determine the virtual computer cluster size needed for the calculation, and provide the virtual computer cluster of the needed size. Calculation parameters such as Monte Carlo calculation parameters are uploaded and the virtual computer cluster size is determined. The calculation parameters may be uploaded to a cloud computing network. It is contemplated the determination of the virtual computer cluster size needed for the calculation is done at the cloud computing network. Virtual calculation nodes may also be requested from the cloud computing network. The virtual computer cluster size may be dynamic. Calculation tasks are distributed and at least one calculation is performed. Calculation results are combined to obtain a final result that can be returned. The calculation results may occur on a single node or one a plurality of nodes in parallel. The final result may be aggregated linearly or by using a Tree-based distribution model.
The cloud computing paradigm disclosed herein provides new opportunities for medical physics. The framework disclosed herein provides an on-demand, virtual computing cluster running on a commercial cloud computing service. The on-demand nature, ease of access, and pay-as-you-go pricing model yield the promise of providing clinics and researchers access to unprecedented amounts of computing power for medical physics calculations.
The cloud computing approach disclosed herein fundamentally alters the economics of such calculations by providing relatively cheap, on-demand, pay-as-you-go computing resources over the Internet. The invention disclosed herein thus enables very large scale Monte Carlo calculations to be routinely performed by clinics and researchers in an efficient and economical way.
The described embodiments are to be considered in all respects only as illustrative and not restrictive, and the scope of the invention is not limited to the foregoing description. Those of skill in the art will recognize changes, substitutions and other modifications that will nonetheless come within the scope of the invention and range of the claims.
DESCRIPTION OF THE DRAWINGS
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT
As described herein, the preferred embodiments of the present invention provide a method for calculating distributions of energy depositions (or dose) and fluence. Described herein are certain embodiments, however, this does not limit other embodiments from using other suitable methods and materials. Those of skill in the art will appreciate that the following description is related to preferred and/or exemplary embodiments of the present invention. Certain embodiments of the present invention are defined exclusively within the appended claims.
Embodiments of the present invention allow a user to calculate distributions of energy deposition (or dose) and fluence due to ionizing radiation using a virtual, on-demand computing cluster over a network such as the Internet instead of using local, dedicated hardware. Such calculations can be used in planning radiation therapy treatments, design and quality control of diagnostic medical imaging, and design or radiation shielding, among other applications.
Cloud computing is a name given to a set of technologies offered as services over the Internet. Cloud providers such as Google, Inc., Amazon Inc., and Rackspace, Inc. offer computing resources, that is, CPU time, storage, etc., which are scalable to the users' needs. Pricing is usually based on a pay-as-you-go model, generally billed on hourly increments, and without set contract periods. This scheme allows cloud services to offer on-demand computing infrastructure, sized to fit the users' momentary needs. Cloud computing has come about because of the commoditization of computer hardware, extensive availability of high bandwidth networks, and growth of free, open source software, including entire operating systems and virtual machine software.
Specifically, the cloud computing system 100 includes at least one client computer 102. The client computer 102 includes memory such as random access memory (“RAM”), read-only memory (“ROM”), mass storage device, or any combination thereof. The memory functions as a computer usable storage medium, otherwise referred to as a computer readable storage medium, to store and/or access computer software and/or instructions.
The client computer 102 also includes a communications interface, for example, a modem, a network interface (such as an Ethernet card), a communications port, a PCMCIA slot and card, wired or wireless systems, etc. The communications interface allows communication through transferred signals between the client computer 102 and external devices including networks such as the Internet 104 and cloud data center 106. Communication may be implemented using wireless or wired capability such as cable, fiber optics, a phone line, a cellular phone link, radio waves or other communication channels.
The client computer 102 establishes communication with the Internet 104—specifically to one or more servers—to, in turn, establish communication with one or more cloud data centers 106. A cloud data center 106 includes one or more networks 110a, 110b, 110c managed through a cloud management system 108. Each network 110a, 110b, 110c includes resource servers 112a, 112b, 112c, respectively. Servers 112a, 112b, 112c permit access to a collection of computing resources and components that can be invoked to instantiate a virtual machine, process, or other resource for a limited or defined duration. For example, one group of resource servers can host and serve an operating system or components thereof to deliver and instantiate a virtual machine. Another group of resource servers can accept requests to host computing cycles or processor time, to supply a defined level of processing power for a virtual machine. A further group of resource servers can host and serve applications to load on an instantiation of a virtual machine, such as an email client, a browser application, a messaging application, a calculations application or other applications or software.
The cloud management system 108 can comprise a dedicated or centralized server and/or other software, hardware, and network tools to communicate with one or more networks 110a, 110b, 110c, such as the Internet or other public or private network, with all sets of resource servers 112a, 112b, 112c. The cloud management system 108 may be configured to query and identify the computing resources and components managed by the set of resource servers 112a, 112b, 112c needed and available for use in the cloud data center 106. Specifically, the cloud management system 108 may be configured to identify the hardware resources and components such as type and amount of processing power, type and amount of memory, type and amount of storage, type and amount of network bandwidth and the like, of the set of resource servers 112a, 112b, 112c needed and available for use in the cloud data center 106. Likewise, the cloud management system 108 can be configured to identify the software resources and components, such as type of Operating System (“OS”), application programs, and the like, of the set of resource servers 112a, 112b, 112c needed and available for use in the cloud data center 106.
The cloud computing system 100 of
An embodiment of the process disclosed herein relies on a cloud computing model of virtual computer resources depicted in
By way of example, such medical physics calculations can be performed through a number of different cloud computing providers. They can be carried out on Amazon Web Service's (AWS) Elastic Compute Cloud (EC2). Several other vendors offer similar cloud services. AWS's EC2 offers several different processor-RAM combinations at different hourly rates. A virtual appliance can be used to instantiate or create a virtual machine within a cloud. To the user, this process is just like booting up a standard server computer. Once the instance is up and running, it will have a unique IP address and a domain name, allowing the user to log in. A virtual cluster can be built by requesting multiple virtual nodes (instances). The size of the cluster can be scaled on demand, that is, virtual nodes can be dynamically created and destroyed. Files can be stored on the running virtual nodes as well as on the persistent storage server at the chosen cloud computing provider to facilitate transfer to and from the cloud. From a user perspective, the persistent storage service can be viewed as the counterpart of the underlying network file system (NFS) found in most cluster computing environments.
According to another exemplary embodiment, a custom distributed processing framework can additionally be implemented to access the chosen cloud computing provider and the secure shell protocol (SSH) for network communications. Such a distributed processing framework can be viewed as a bridge between the clinics and commercial cloud computing providers. Hospitals and clinics can interact with the distributed processing framework through a custom designed platform interface. The distributed processing framework can host such functions as keeping track of the available nodes in a virtual cluster, launching the requisite number of nodes by, dynamically monitoring the size of the cluster, and collecting computation results. As such, at any moment, the distributed processing framework can maintain a virtual cluster and dynamically monitor the health of the cluster. For example, if some nodes failed, it would immediately request replacements from the cloud; if system load were low, it would dynamically reduce the size of the cluster. The use of the virtual nodes and other computing resources can be administered by a scheduling function residing on the distributed processing framework. The scheduling function helps to keep track of the available nodes in the virtual cluster. The job launching function determines the number of nodes needed for the computation based on the time constraints, the anatomy and the number of particles to be simulated and request the appropriate number of nodes from the scheduling function. Since nodes may fail, redundancy is preferably incorporated. For example, instead of requesting n number of nodes for the computation, the job launching function may request 5% more nodes, or some threshold based on reliability statistics. The results collecting function will collect the results from the virtual calculation nodes
The general data flow of the cloud based calculations incorporating the distributed processing framework is discussed in detail with reference to
A Monte Carlo calculation task (including patient anatomy, beam configuration, etc.) is uploaded onto the distributed processing framework 320 using the secure platform interface 302 via the Internet 310 at step 410. At step 420 the job launching function 323 determines the number of nodes needed for the computation based on the time-constraint, the anatomy and the number of primaries needed. At step 430 the job launching function requests the appropriate number of nodes from the scheduling function 324. Since the nodes may fail, redundancy may be incorporated. At step 440 the Monte Carlo simulation is launched on each requested node in a distributive manner. Extra care is used for random seed generations to ensure the correctness of the simulations. At step 450 the progress of the computation is dynamically monitored. Since nodes may fail during the computation, success is declared once the number of nodes that completes the computation reaches a certain threshold, for example 95% of nodes. The cloud provider interface 335 may be in communication with the distributed processing framework 320.
Once the computation is finished on each of the virtual nodes, the results are aggregated at step 460 using a linear model or a tree-based distributive model, and a final dose distribution is returned to the user.
Various exemplary embodiments have been described above. Those skilled in the art will understand, however, that changes and modifications may be made to those examples without departing from the scope of the claims.
1. A method of determining distributions of energy deposition and fluence due to ionizing radiation using a virtual computer cluster over a network comprising the steps of:
- uploading calculation parameters;
- determining virtual computer cluster size;
- distributing calculation tasks;
- performing at least one calculation;
- combining calculation results from at least one calculation to obtain a final result; and
- returning the final result.
2. The method of claim 1, wherein the virtual computer cluster size is dynamic.
3. The method of claim 1, wherein the final result is aggregated linearly.
4. The method of claim 1, where in the final result is aggregated using a tree-based distribution model.
5. The method of claim 1, wherein the calculation results occur on a single node.
6. The method of claim 1, further comprising the step of accessing a distributed processing framework.
7. The method of claim 6, wherein the distributed processing framework dynamically monitors the virtual computer cluster.
8. The method of claim 6, wherein the distributed processing framework determines the virtual computer cluster size needed for the calculation.
9. The method of claim 6, wherein the distributed processing framework provides the virtual computer cluster of the needed size.
10. The method of claim 1, wherein calculation parameters are Monte Carlo calculation parameters.
11. The method of claim 1, wherein calculation parameters are uploaded to a cloud computing network.
12. The method of claim 1, determining the virtual computer cluster size needed for the calculation is done at the cloud computing network.
13. The method of claim 1, further comprising the step of requesting virtual calculation nodes from a cloud computing network.
14. A method of determining distributions of energy deposition and fluence due to ionizing radiation using a virtual computer cluster over a network comprising the steps of:
- uploading calculation parameters to a distributed processing framework;
- determining virtual computer cluster size;
- dynamically requesting virtual calculation nodes from a cloud computing network;
- receiving a dynamic set of virtual calculation nodes;
- distributing calculation tasks;
- performing at least one calculation;
- combining calculation results from at least one calculation to obtain a final result; and
- returning the final result.
15. The method of claim 14, wherein the step of dynamically requesting virtual calculation nodes from a cloud computing network and the step of receiving a dynamic set of virtual calculation nodes are done at least once before proceeding to the step of distributing calculation tasks.
Filed: Apr 28, 2011
Publication Date: Feb 28, 2013
Patent Grant number: 9805170
Inventors: Roy William Keyes (Albuquerque, NM), Christian Romano (Albuquerque, NM), Shuang Luan (Albuquerque, NM), Dorian C. Arnold (Albuquerque, NM)
Application Number: 13/643,844
International Classification: G06F 19/00 (20110101); G06F 15/16 (20060101);